2 This file is part of systemd.
4 Copyright 2011 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/statvfs.h>
30 #include "alloc-util.h"
31 #include "btrfs-util.h"
32 #include "chattr-util.h"
35 #include "journal-authenticate.h"
36 #include "journal-def.h"
37 #include "journal-file.h"
39 #include "parse-util.h"
40 #include "path-util.h"
41 #include "random-util.h"
44 #include "string-util.h"
45 #include "xattr-util.h"
47 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
48 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
50 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
52 /* This is the minimum journal file size */
53 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
55 /* These are the lower and upper bounds if we deduce the max_use value
56 * from the file system size */
57 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
58 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
60 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
61 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
63 /* This is the upper bound if we deduce max_size from max_use */
64 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
66 /* This is the upper bound if we deduce the keep_free value from the
68 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
70 /* This is the keep_free value when we can't determine the system
72 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
74 /* This is the default maximum number of journal files to keep around. */
75 #define DEFAULT_N_MAX_FILES (100)
77 /* n_data was the first entry we added after the initial file format design */
78 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
80 /* How many entries to keep in the entry array chain cache at max */
81 #define CHAIN_CACHE_MAX 20
83 /* How much to increase the journal file size at once each time we allocate something new. */
84 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
86 /* Reread fstat() of the file for detecting deletions at least this often */
87 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
89 /* The mmap context to use for the header we pick as one above the last defined typed */
90 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
92 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
93 * As a result we use atomic operations on f->offline_state for inter-thread communications with
94 * journal_file_set_offline() and journal_file_set_online(). */
95 static void journal_file_set_offline_internal(JournalFile
*f
) {
101 switch (f
->offline_state
) {
103 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
107 case OFFLINE_AGAIN_FROM_SYNCING
:
108 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
112 case OFFLINE_AGAIN_FROM_OFFLINING
:
113 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
117 case OFFLINE_SYNCING
:
120 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
123 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
127 case OFFLINE_OFFLINING
:
128 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
136 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
142 static void * journal_file_set_offline_thread(void *arg
) {
143 JournalFile
*f
= arg
;
145 journal_file_set_offline_internal(f
);
150 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
155 if (f
->offline_state
== OFFLINE_JOINED
)
158 r
= pthread_join(f
->offline_thread
, NULL
);
162 f
->offline_state
= OFFLINE_JOINED
;
164 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
170 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
171 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
173 switch (f
->offline_state
) {
174 case OFFLINE_AGAIN_FROM_SYNCING
:
175 case OFFLINE_AGAIN_FROM_OFFLINING
:
179 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
183 case OFFLINE_SYNCING
:
184 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
188 case OFFLINE_OFFLINING
:
189 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
199 /* Sets a journal offline.
201 * If wait is false then an offline is dispatched in a separate thread for a
202 * subsequent journal_file_set_offline() or journal_file_set_online() of the
203 * same journal to synchronize with.
205 * If wait is true, then either an existing offline thread will be restarted
206 * and joined, or if none exists the offline is simply performed in this
207 * context without involving another thread.
209 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
218 if (!(f
->fd
>= 0 && f
->header
))
221 /* An offlining journal is implicitly online and may modify f->header->state,
222 * we must also join any potentially lingering offline thread when not online. */
223 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
224 return journal_file_set_offline_thread_join(f
);
226 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
227 restarted
= journal_file_set_offline_try_restart(f
);
228 if ((restarted
&& wait
) || !restarted
) {
229 r
= journal_file_set_offline_thread_join(f
);
237 /* Initiate a new offline. */
238 f
->offline_state
= OFFLINE_SYNCING
;
240 if (wait
) /* Without using a thread if waiting. */
241 journal_file_set_offline_internal(f
);
243 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
245 f
->offline_state
= OFFLINE_JOINED
;
253 static int journal_file_set_online(JournalFile
*f
) {
261 if (!(f
->fd
>= 0 && f
->header
))
265 switch (f
->offline_state
) {
267 /* No offline thread, no need to wait. */
271 case OFFLINE_SYNCING
:
272 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
274 /* Canceled syncing prior to offlining, no need to wait. */
277 case OFFLINE_AGAIN_FROM_SYNCING
:
278 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
280 /* Canceled restart from syncing, no need to wait. */
283 case OFFLINE_AGAIN_FROM_OFFLINING
:
284 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
286 /* Canceled restart from offlining, must wait for offlining to complete however. */
288 /* fall through to wait */
292 r
= journal_file_set_offline_thread_join(f
);
302 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
305 switch (f
->header
->state
) {
310 f
->header
->state
= STATE_ONLINE
;
319 bool journal_file_is_offlining(JournalFile
*f
) {
322 __sync_synchronize();
324 if (f
->offline_state
== OFFLINE_DONE
||
325 f
->offline_state
== OFFLINE_JOINED
)
331 JournalFile
* journal_file_close(JournalFile
*f
) {
335 /* Write the final tag */
336 if (f
->seal
&& f
->writable
) {
339 r
= journal_file_append_tag(f
);
341 log_error_errno(r
, "Failed to append tag when closing journal: %m");
345 if (f
->post_change_timer
) {
348 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
349 if (enabled
== SD_EVENT_ONESHOT
)
350 journal_file_post_change(f
);
352 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
353 sd_event_source_unref(f
->post_change_timer
);
356 journal_file_set_offline(f
, true);
358 if (f
->mmap
&& f
->fd
>= 0)
359 mmap_cache_close_fd(f
->mmap
, f
->fd
);
361 if (f
->fd
>= 0 && f
->defrag_on_close
) {
363 /* Be friendly to btrfs: turn COW back on again now,
364 * and defragment the file. We won't write to the file
365 * ever again, hence remove all fragmentation, and
366 * reenable all the good bits COW usually provides
367 * (such as data checksumming). */
369 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
370 (void) btrfs_defrag_fd(f
->fd
);
377 mmap_cache_unref(f
->mmap
);
379 ordered_hashmap_free_free(f
->chain_cache
);
381 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
382 free(f
->compress_buffer
);
387 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
389 free(f
->fsprg_state
);
394 gcry_md_close(f
->hmac
);
400 void journal_file_close_set(Set
*s
) {
405 while ((f
= set_steal_first(s
)))
406 (void) journal_file_close(f
);
409 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
416 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
417 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
419 h
.incompatible_flags
|= htole32(
420 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
421 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
423 h
.compatible_flags
= htole32(
424 f
->seal
* HEADER_COMPATIBLE_SEALED
);
426 r
= sd_id128_randomize(&h
.file_id
);
431 h
.seqnum_id
= template->header
->seqnum_id
;
432 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
434 h
.seqnum_id
= h
.file_id
;
436 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
446 static int fsync_directory_of_file(int fd
) {
447 _cleanup_free_
char *path
= NULL
, *dn
= NULL
;
448 _cleanup_close_
int dfd
= -1;
452 if (fstat(fd
, &st
) < 0)
455 if (!S_ISREG(st
.st_mode
))
458 r
= fd_get_path(fd
, &path
);
462 if (!path_is_absolute(path
))
465 dn
= dirname_malloc(path
);
469 dfd
= open(dn
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
);
479 static int journal_file_refresh_header(JournalFile
*f
) {
486 r
= sd_id128_get_machine(&f
->header
->machine_id
);
490 r
= sd_id128_get_boot(&boot_id
);
494 if (sd_id128_equal(boot_id
, f
->header
->boot_id
))
495 f
->tail_entry_monotonic_valid
= true;
497 f
->header
->boot_id
= boot_id
;
499 r
= journal_file_set_online(f
);
501 /* Sync the online state to disk */
504 /* We likely just created a new file, also sync the directory this file is located in. */
505 (void) fsync_directory_of_file(f
->fd
);
510 static int journal_file_verify_header(JournalFile
*f
) {
516 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
519 /* In both read and write mode we refuse to open files with
520 * incompatible flags we don't know */
521 flags
= le32toh(f
->header
->incompatible_flags
);
522 if (flags
& ~HEADER_INCOMPATIBLE_SUPPORTED
) {
523 if (flags
& ~HEADER_INCOMPATIBLE_ANY
)
524 log_debug("Journal file %s has unknown incompatible flags %"PRIx32
,
525 f
->path
, flags
& ~HEADER_INCOMPATIBLE_ANY
);
526 flags
= (flags
& HEADER_INCOMPATIBLE_ANY
) & ~HEADER_INCOMPATIBLE_SUPPORTED
;
528 log_debug("Journal file %s uses incompatible flags %"PRIx32
529 " disabled at compilation time.", f
->path
, flags
);
530 return -EPROTONOSUPPORT
;
533 /* When open for writing we refuse to open files with
534 * compatible flags, too */
535 flags
= le32toh(f
->header
->compatible_flags
);
536 if (f
->writable
&& (flags
& ~HEADER_COMPATIBLE_SUPPORTED
)) {
537 if (flags
& ~HEADER_COMPATIBLE_ANY
)
538 log_debug("Journal file %s has unknown compatible flags %"PRIx32
,
539 f
->path
, flags
& ~HEADER_COMPATIBLE_ANY
);
540 flags
= (flags
& HEADER_COMPATIBLE_ANY
) & ~HEADER_COMPATIBLE_SUPPORTED
;
542 log_debug("Journal file %s uses compatible flags %"PRIx32
543 " disabled at compilation time.", f
->path
, flags
);
544 return -EPROTONOSUPPORT
;
547 if (f
->header
->state
>= _STATE_MAX
)
550 /* The first addition was n_data, so check that we are at least this large */
551 if (le64toh(f
->header
->header_size
) < HEADER_SIZE_MIN
)
554 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
557 if ((le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)) > (uint64_t) f
->last_stat
.st_size
)
560 if (le64toh(f
->header
->tail_object_offset
) > (le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)))
563 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
564 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
565 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
566 !VALID64(le64toh(f
->header
->entry_array_offset
)))
570 sd_id128_t machine_id
;
574 r
= sd_id128_get_machine(&machine_id
);
578 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
581 state
= f
->header
->state
;
583 if (state
== STATE_ONLINE
) {
584 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
586 } else if (state
== STATE_ARCHIVED
)
588 else if (state
!= STATE_OFFLINE
) {
589 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
593 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
594 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
596 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
597 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
602 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
603 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
605 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
610 static int journal_file_fstat(JournalFile
*f
) {
614 if (fstat(f
->fd
, &f
->last_stat
) < 0)
617 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
619 /* Refuse appending to files that are already deleted */
620 if (f
->last_stat
.st_nlink
<= 0)
626 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
627 uint64_t old_size
, new_size
;
633 /* We assume that this file is not sparse, and we know that
634 * for sure, since we always call posix_fallocate()
637 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
641 le64toh(f
->header
->header_size
) +
642 le64toh(f
->header
->arena_size
);
644 new_size
= PAGE_ALIGN(offset
+ size
);
645 if (new_size
< le64toh(f
->header
->header_size
))
646 new_size
= le64toh(f
->header
->header_size
);
648 if (new_size
<= old_size
) {
650 /* We already pre-allocated enough space, but before
651 * we write to it, let's check with fstat() if the
652 * file got deleted, in order make sure we don't throw
653 * away the data immediately. Don't check fstat() for
654 * all writes though, but only once ever 10s. */
656 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
659 return journal_file_fstat(f
);
662 /* Allocate more space. */
664 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
667 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
670 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
673 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
675 if (new_size
- old_size
> available
)
680 /* Increase by larger blocks at once */
681 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
682 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
683 new_size
= f
->metrics
.max_size
;
685 /* Note that the glibc fallocate() fallback is very
686 inefficient, hence we try to minimize the allocation area
688 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
692 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
694 return journal_file_fstat(f
);
697 static unsigned type_to_context(ObjectType type
) {
698 /* One context for each type, plus one catch-all for the rest */
699 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
700 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
701 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
704 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
) {
713 /* Avoid SIGBUS on invalid accesses */
714 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
715 /* Hmm, out of range? Let's refresh the fstat() data
716 * first, before we trust that check. */
718 r
= journal_file_fstat(f
);
722 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
723 return -EADDRNOTAVAIL
;
726 return mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
729 static uint64_t minimum_header_size(Object
*o
) {
731 static const uint64_t table
[] = {
732 [OBJECT_DATA
] = sizeof(DataObject
),
733 [OBJECT_FIELD
] = sizeof(FieldObject
),
734 [OBJECT_ENTRY
] = sizeof(EntryObject
),
735 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
736 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
737 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
738 [OBJECT_TAG
] = sizeof(TagObject
),
741 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
742 return sizeof(ObjectHeader
);
744 return table
[o
->object
.type
];
747 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
756 /* Objects may only be located at multiple of 64 bit */
757 if (!VALID64(offset
)) {
758 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
762 /* Object may not be located in the file header */
763 if (offset
< le64toh(f
->header
->header_size
)) {
764 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
768 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
);
773 s
= le64toh(o
->object
.size
);
776 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
779 if (s
< sizeof(ObjectHeader
)) {
780 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
784 if (o
->object
.type
<= OBJECT_UNUSED
) {
785 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
789 if (s
< minimum_header_size(o
)) {
790 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
794 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
795 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
799 if (s
> sizeof(ObjectHeader
)) {
800 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
);
811 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
817 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
820 /* If an external seqnum counter was passed, we update
821 * both the local and the external one, and set it to
822 * the maximum of both */
830 f
->header
->tail_entry_seqnum
= htole64(r
);
832 if (f
->header
->head_entry_seqnum
== 0)
833 f
->header
->head_entry_seqnum
= htole64(r
);
838 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
846 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
847 assert(size
>= sizeof(ObjectHeader
));
851 r
= journal_file_set_online(f
);
855 p
= le64toh(f
->header
->tail_object_offset
);
857 p
= le64toh(f
->header
->header_size
);
859 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
863 p
+= ALIGN64(le64toh(tail
->object
.size
));
866 r
= journal_file_allocate(f
, p
, size
);
870 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
);
877 o
->object
.type
= type
;
878 o
->object
.size
= htole64(size
);
880 f
->header
->tail_object_offset
= htole64(p
);
881 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
889 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
897 /* We estimate that we need 1 hash table entry per 768 bytes
898 of journal file and we want to make sure we never get
899 beyond 75% fill level. Calculate the hash table size for
900 the maximum file size based on these metrics. */
902 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
903 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
904 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
906 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
908 r
= journal_file_append_object(f
,
909 OBJECT_DATA_HASH_TABLE
,
910 offsetof(Object
, hash_table
.items
) + s
,
915 memzero(o
->hash_table
.items
, s
);
917 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
918 f
->header
->data_hash_table_size
= htole64(s
);
923 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
931 /* We use a fixed size hash table for the fields as this
932 * number should grow very slowly only */
934 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
935 r
= journal_file_append_object(f
,
936 OBJECT_FIELD_HASH_TABLE
,
937 offsetof(Object
, hash_table
.items
) + s
,
942 memzero(o
->hash_table
.items
, s
);
944 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
945 f
->header
->field_hash_table_size
= htole64(s
);
950 int journal_file_map_data_hash_table(JournalFile
*f
) {
958 if (f
->data_hash_table
)
961 p
= le64toh(f
->header
->data_hash_table_offset
);
962 s
= le64toh(f
->header
->data_hash_table_size
);
964 r
= journal_file_move_to(f
,
965 OBJECT_DATA_HASH_TABLE
,
972 f
->data_hash_table
= t
;
976 int journal_file_map_field_hash_table(JournalFile
*f
) {
984 if (f
->field_hash_table
)
987 p
= le64toh(f
->header
->field_hash_table_offset
);
988 s
= le64toh(f
->header
->field_hash_table_size
);
990 r
= journal_file_move_to(f
,
991 OBJECT_FIELD_HASH_TABLE
,
998 f
->field_hash_table
= t
;
1002 static int journal_file_link_field(
1013 assert(f
->field_hash_table
);
1017 if (o
->object
.type
!= OBJECT_FIELD
)
1020 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1024 /* This might alter the window we are looking at */
1025 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1028 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1030 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1032 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1036 o
->field
.next_hash_offset
= htole64(offset
);
1039 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1041 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1042 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1047 static int journal_file_link_data(
1058 assert(f
->data_hash_table
);
1062 if (o
->object
.type
!= OBJECT_DATA
)
1065 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1069 /* This might alter the window we are looking at */
1070 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1071 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1072 o
->data
.n_entries
= 0;
1075 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1077 /* Only entry in the hash table is easy */
1078 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1080 /* Move back to the previous data object, to patch in
1083 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1087 o
->data
.next_hash_offset
= htole64(offset
);
1090 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1092 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1093 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1098 int journal_file_find_field_object_with_hash(
1100 const void *field
, uint64_t size
, uint64_t hash
,
1101 Object
**ret
, uint64_t *offset
) {
1103 uint64_t p
, osize
, h
, m
;
1108 assert(field
&& size
> 0);
1110 /* If the field hash table is empty, we can't find anything */
1111 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1114 /* Map the field hash table, if it isn't mapped yet. */
1115 r
= journal_file_map_field_hash_table(f
);
1119 osize
= offsetof(Object
, field
.payload
) + size
;
1121 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1126 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1131 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1135 if (le64toh(o
->field
.hash
) == hash
&&
1136 le64toh(o
->object
.size
) == osize
&&
1137 memcmp(o
->field
.payload
, field
, size
) == 0) {
1147 p
= le64toh(o
->field
.next_hash_offset
);
1153 int journal_file_find_field_object(
1155 const void *field
, uint64_t size
,
1156 Object
**ret
, uint64_t *offset
) {
1161 assert(field
&& size
> 0);
1163 hash
= hash64(field
, size
);
1165 return journal_file_find_field_object_with_hash(f
,
1170 int journal_file_find_data_object_with_hash(
1172 const void *data
, uint64_t size
, uint64_t hash
,
1173 Object
**ret
, uint64_t *offset
) {
1175 uint64_t p
, osize
, h
, m
;
1180 assert(data
|| size
== 0);
1182 /* If there's no data hash table, then there's no entry. */
1183 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1186 /* Map the data hash table, if it isn't mapped yet. */
1187 r
= journal_file_map_data_hash_table(f
);
1191 osize
= offsetof(Object
, data
.payload
) + size
;
1193 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1198 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1203 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1207 if (le64toh(o
->data
.hash
) != hash
)
1210 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1211 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1215 l
= le64toh(o
->object
.size
);
1216 if (l
<= offsetof(Object
, data
.payload
))
1219 l
-= offsetof(Object
, data
.payload
);
1221 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1222 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1226 if (rsize
== size
&&
1227 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1238 return -EPROTONOSUPPORT
;
1240 } else if (le64toh(o
->object
.size
) == osize
&&
1241 memcmp(o
->data
.payload
, data
, size
) == 0) {
1253 p
= le64toh(o
->data
.next_hash_offset
);
1259 int journal_file_find_data_object(
1261 const void *data
, uint64_t size
,
1262 Object
**ret
, uint64_t *offset
) {
1267 assert(data
|| size
== 0);
1269 hash
= hash64(data
, size
);
1271 return journal_file_find_data_object_with_hash(f
,
1276 static int journal_file_append_field(
1278 const void *field
, uint64_t size
,
1279 Object
**ret
, uint64_t *offset
) {
1287 assert(field
&& size
> 0);
1289 hash
= hash64(field
, size
);
1291 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1305 osize
= offsetof(Object
, field
.payload
) + size
;
1306 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1310 o
->field
.hash
= htole64(hash
);
1311 memcpy(o
->field
.payload
, field
, size
);
1313 r
= journal_file_link_field(f
, o
, p
, hash
);
1317 /* The linking might have altered the window, so let's
1318 * refresh our pointer */
1319 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1324 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1338 static int journal_file_append_data(
1340 const void *data
, uint64_t size
,
1341 Object
**ret
, uint64_t *offset
) {
1346 int r
, compression
= 0;
1350 assert(data
|| size
== 0);
1352 hash
= hash64(data
, size
);
1354 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1368 osize
= offsetof(Object
, data
.payload
) + size
;
1369 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1373 o
->data
.hash
= htole64(hash
);
1375 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1376 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1379 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1381 if (compression
>= 0) {
1382 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1383 o
->object
.flags
|= compression
;
1385 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1386 size
, rsize
, object_compressed_to_string(compression
));
1388 /* Compression didn't work, we don't really care why, let's continue without compression */
1393 if (compression
== 0)
1394 memcpy_safe(o
->data
.payload
, data
, size
);
1396 r
= journal_file_link_data(f
, o
, p
, hash
);
1401 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1406 /* The linking might have altered the window, so let's
1407 * refresh our pointer */
1408 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1415 eq
= memchr(data
, '=', size
);
1416 if (eq
&& eq
> data
) {
1420 /* Create field object ... */
1421 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1425 /* ... and link it in. */
1426 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1427 fo
->field
.head_data_offset
= le64toh(p
);
1439 uint64_t journal_file_entry_n_items(Object
*o
) {
1442 if (o
->object
.type
!= OBJECT_ENTRY
)
1445 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1448 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1451 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1454 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1457 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1460 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1461 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1464 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1467 static int link_entry_into_array(JournalFile
*f
,
1472 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1481 a
= le64toh(*first
);
1482 i
= hidx
= le64toh(*idx
);
1485 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1489 n
= journal_file_entry_array_n_items(o
);
1491 o
->entry_array
.items
[i
] = htole64(p
);
1492 *idx
= htole64(hidx
+ 1);
1498 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1509 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1510 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1516 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1521 o
->entry_array
.items
[i
] = htole64(p
);
1524 *first
= htole64(q
);
1526 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1530 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1533 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1534 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1536 *idx
= htole64(hidx
+ 1);
1541 static int link_entry_into_array_plus_one(JournalFile
*f
,
1556 *extra
= htole64(p
);
1560 i
= htole64(le64toh(*idx
) - 1);
1561 r
= link_entry_into_array(f
, first
, &i
, p
);
1566 *idx
= htole64(le64toh(*idx
) + 1);
1570 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1577 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1581 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1585 return link_entry_into_array_plus_one(f
,
1586 &o
->data
.entry_offset
,
1587 &o
->data
.entry_array_offset
,
1592 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1601 if (o
->object
.type
!= OBJECT_ENTRY
)
1604 __sync_synchronize();
1606 /* Link up the entry itself */
1607 r
= link_entry_into_array(f
,
1608 &f
->header
->entry_array_offset
,
1609 &f
->header
->n_entries
,
1614 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1616 if (f
->header
->head_entry_realtime
== 0)
1617 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1619 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1620 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1622 f
->tail_entry_monotonic_valid
= true;
1624 /* Link up the items */
1625 n
= journal_file_entry_n_items(o
);
1626 for (i
= 0; i
< n
; i
++) {
1627 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1635 static int journal_file_append_entry_internal(
1637 const dual_timestamp
*ts
,
1639 const EntryItem items
[], unsigned n_items
,
1641 Object
**ret
, uint64_t *offset
) {
1649 assert(items
|| n_items
== 0);
1652 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1654 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1658 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1659 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1660 o
->entry
.realtime
= htole64(ts
->realtime
);
1661 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1662 o
->entry
.xor_hash
= htole64(xor_hash
);
1663 o
->entry
.boot_id
= f
->header
->boot_id
;
1666 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1671 r
= journal_file_link_entry(f
, o
, np
);
1684 void journal_file_post_change(JournalFile
*f
) {
1687 /* inotify() does not receive IN_MODIFY events from file
1688 * accesses done via mmap(). After each access we hence
1689 * trigger IN_MODIFY by truncating the journal file to its
1690 * current size which triggers IN_MODIFY. */
1692 __sync_synchronize();
1694 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1695 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1698 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1701 journal_file_post_change(userdata
);
1706 static void schedule_post_change(JournalFile
*f
) {
1707 sd_event_source
*timer
;
1712 assert(f
->post_change_timer
);
1714 timer
= f
->post_change_timer
;
1716 r
= sd_event_source_get_enabled(timer
, &enabled
);
1718 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1722 if (enabled
== SD_EVENT_ONESHOT
)
1725 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1727 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1731 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1733 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1737 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1739 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1746 /* On failure, let's simply post the change immediately. */
1747 journal_file_post_change(f
);
1750 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1751 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1752 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1756 assert_return(!f
->post_change_timer
, -EINVAL
);
1760 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1764 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1768 f
->post_change_timer
= timer
;
1770 f
->post_change_timer_period
= t
;
1775 static int entry_item_cmp(const void *_a
, const void *_b
) {
1776 const EntryItem
*a
= _a
, *b
= _b
;
1778 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1780 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1785 int journal_file_append_entry(JournalFile
*f
, const dual_timestamp
*ts
, const struct iovec iovec
[], unsigned n_iovec
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
1789 uint64_t xor_hash
= 0;
1790 struct dual_timestamp _ts
;
1794 assert(iovec
|| n_iovec
== 0);
1797 dual_timestamp_get(&_ts
);
1802 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1807 /* alloca() can't take 0, hence let's allocate at least one */
1808 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
1810 for (i
= 0; i
< n_iovec
; i
++) {
1814 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1818 xor_hash
^= le64toh(o
->data
.hash
);
1819 items
[i
].object_offset
= htole64(p
);
1820 items
[i
].hash
= o
->data
.hash
;
1823 /* Order by the position on disk, in order to improve seek
1824 * times for rotating media. */
1825 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1827 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1829 /* If the memory mapping triggered a SIGBUS then we return an
1830 * IO error and ignore the error code passed down to us, since
1831 * it is very likely just an effect of a nullified replacement
1834 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
1837 if (f
->post_change_timer
)
1838 schedule_post_change(f
);
1840 journal_file_post_change(f
);
1845 typedef struct ChainCacheItem
{
1846 uint64_t first
; /* the array at the beginning of the chain */
1847 uint64_t array
; /* the cached array */
1848 uint64_t begin
; /* the first item in the cached array */
1849 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1850 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1853 static void chain_cache_put(
1860 uint64_t last_index
) {
1863 /* If the chain item to cache for this chain is the
1864 * first one it's not worth caching anything */
1868 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
1869 ci
= ordered_hashmap_steal_first(h
);
1872 ci
= new(ChainCacheItem
, 1);
1879 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
1884 assert(ci
->first
== first
);
1889 ci
->last_index
= last_index
;
1892 static int generic_array_get(
1896 Object
**ret
, uint64_t *offset
) {
1899 uint64_t p
= 0, a
, t
= 0;
1907 /* Try the chain cache first */
1908 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1909 if (ci
&& i
> ci
->total
) {
1918 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1922 k
= journal_file_entry_array_n_items(o
);
1924 p
= le64toh(o
->entry_array
.items
[i
]);
1930 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1936 /* Let's cache this item for the next invocation */
1937 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
1939 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
1952 static int generic_array_get_plus_one(
1957 Object
**ret
, uint64_t *offset
) {
1966 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
1979 return generic_array_get(f
, first
, i
-1, ret
, offset
);
1988 static int generic_array_bisect(
1993 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
1994 direction_t direction
,
1999 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2000 bool subtract_one
= false;
2001 Object
*o
, *array
= NULL
;
2006 assert(test_object
);
2008 /* Start with the first array in the chain */
2011 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2012 if (ci
&& n
> ci
->total
) {
2013 /* Ah, we have iterated this bisection array chain
2014 * previously! Let's see if we can skip ahead in the
2015 * chain, as far as the last time. But we can't jump
2016 * backwards in the chain, so let's check that
2019 r
= test_object(f
, ci
->begin
, needle
);
2023 if (r
== TEST_LEFT
) {
2024 /* OK, what we are looking for is right of the
2025 * begin of this EntryArray, so let's jump
2026 * straight to previously cached array in the
2032 last_index
= ci
->last_index
;
2037 uint64_t left
, right
, k
, lp
;
2039 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2043 k
= journal_file_entry_array_n_items(array
);
2049 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2053 r
= test_object(f
, p
, needle
);
2054 if (r
== -EBADMSG
) {
2055 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2062 if (r
== TEST_FOUND
)
2063 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2065 if (r
== TEST_RIGHT
) {
2069 if (last_index
!= (uint64_t) -1) {
2070 assert(last_index
<= right
);
2072 /* If we cached the last index we
2073 * looked at, let's try to not to jump
2074 * too wildly around and see if we can
2075 * limit the range to look at early to
2076 * the immediate neighbors of the last
2077 * index we looked at. */
2079 if (last_index
> 0) {
2080 uint64_t x
= last_index
- 1;
2082 p
= le64toh(array
->entry_array
.items
[x
]);
2086 r
= test_object(f
, p
, needle
);
2090 if (r
== TEST_FOUND
)
2091 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2093 if (r
== TEST_RIGHT
)
2099 if (last_index
< right
) {
2100 uint64_t y
= last_index
+ 1;
2102 p
= le64toh(array
->entry_array
.items
[y
]);
2106 r
= test_object(f
, p
, needle
);
2110 if (r
== TEST_FOUND
)
2111 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2113 if (r
== TEST_RIGHT
)
2121 if (left
== right
) {
2122 if (direction
== DIRECTION_UP
)
2123 subtract_one
= true;
2129 assert(left
< right
);
2130 i
= (left
+ right
) / 2;
2132 p
= le64toh(array
->entry_array
.items
[i
]);
2136 r
= test_object(f
, p
, needle
);
2137 if (r
== -EBADMSG
) {
2138 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2145 if (r
== TEST_FOUND
)
2146 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2148 if (r
== TEST_RIGHT
)
2156 if (direction
== DIRECTION_UP
) {
2158 subtract_one
= true;
2169 last_index
= (uint64_t) -1;
2170 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2176 if (subtract_one
&& t
== 0 && i
== 0)
2179 /* Let's cache this item for the next invocation */
2180 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(array
->entry_array
.items
[0]), t
, subtract_one
? (i
> 0 ? i
-1 : (uint64_t) -1) : i
);
2182 if (subtract_one
&& i
== 0)
2184 else if (subtract_one
)
2185 p
= le64toh(array
->entry_array
.items
[i
-1]);
2187 p
= le64toh(array
->entry_array
.items
[i
]);
2189 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2200 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2205 static int generic_array_bisect_plus_one(
2211 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2212 direction_t direction
,
2218 bool step_back
= false;
2222 assert(test_object
);
2227 /* This bisects the array in object 'first', but first checks
2229 r
= test_object(f
, extra
, needle
);
2233 if (r
== TEST_FOUND
)
2234 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2236 /* if we are looking with DIRECTION_UP then we need to first
2237 see if in the actual array there is a matching entry, and
2238 return the last one of that. But if there isn't any we need
2239 to return this one. Hence remember this, and return it
2242 step_back
= direction
== DIRECTION_UP
;
2244 if (r
== TEST_RIGHT
) {
2245 if (direction
== DIRECTION_DOWN
)
2251 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2253 if (r
== 0 && step_back
)
2262 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2278 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2284 else if (p
< needle
)
2290 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2297 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2301 if (le64toh(o
->entry
.seqnum
) == needle
)
2303 else if (le64toh(o
->entry
.seqnum
) < needle
)
2309 int journal_file_move_to_entry_by_seqnum(
2312 direction_t direction
,
2318 return generic_array_bisect(f
,
2319 le64toh(f
->header
->entry_array_offset
),
2320 le64toh(f
->header
->n_entries
),
2327 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2334 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2338 if (le64toh(o
->entry
.realtime
) == needle
)
2340 else if (le64toh(o
->entry
.realtime
) < needle
)
2346 int journal_file_move_to_entry_by_realtime(
2349 direction_t direction
,
2355 return generic_array_bisect(f
,
2356 le64toh(f
->header
->entry_array_offset
),
2357 le64toh(f
->header
->n_entries
),
2359 test_object_realtime
,
2364 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2371 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2375 if (le64toh(o
->entry
.monotonic
) == needle
)
2377 else if (le64toh(o
->entry
.monotonic
) < needle
)
2383 static int find_data_object_by_boot_id(
2389 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2391 sd_id128_to_string(boot_id
, t
+ 9);
2392 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2395 int journal_file_move_to_entry_by_monotonic(
2399 direction_t direction
,
2408 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2414 return generic_array_bisect_plus_one(f
,
2415 le64toh(o
->data
.entry_offset
),
2416 le64toh(o
->data
.entry_array_offset
),
2417 le64toh(o
->data
.n_entries
),
2419 test_object_monotonic
,
2424 void journal_file_reset_location(JournalFile
*f
) {
2425 f
->location_type
= LOCATION_HEAD
;
2426 f
->current_offset
= 0;
2427 f
->current_seqnum
= 0;
2428 f
->current_realtime
= 0;
2429 f
->current_monotonic
= 0;
2430 zero(f
->current_boot_id
);
2431 f
->current_xor_hash
= 0;
2434 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2435 f
->location_type
= LOCATION_SEEK
;
2436 f
->current_offset
= offset
;
2437 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2438 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2439 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2440 f
->current_boot_id
= o
->entry
.boot_id
;
2441 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2444 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2449 assert(af
->location_type
== LOCATION_SEEK
);
2450 assert(bf
->location_type
== LOCATION_SEEK
);
2452 /* If contents and timestamps match, these entries are
2453 * identical, even if the seqnum does not match */
2454 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2455 af
->current_monotonic
== bf
->current_monotonic
&&
2456 af
->current_realtime
== bf
->current_realtime
&&
2457 af
->current_xor_hash
== bf
->current_xor_hash
)
2460 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2462 /* If this is from the same seqnum source, compare
2464 if (af
->current_seqnum
< bf
->current_seqnum
)
2466 if (af
->current_seqnum
> bf
->current_seqnum
)
2469 /* Wow! This is weird, different data but the same
2470 * seqnums? Something is borked, but let's make the
2471 * best of it and compare by time. */
2474 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2476 /* If the boot id matches, compare monotonic time */
2477 if (af
->current_monotonic
< bf
->current_monotonic
)
2479 if (af
->current_monotonic
> bf
->current_monotonic
)
2483 /* Otherwise, compare UTC time */
2484 if (af
->current_realtime
< bf
->current_realtime
)
2486 if (af
->current_realtime
> bf
->current_realtime
)
2489 /* Finally, compare by contents */
2490 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2492 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2498 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2500 /* Increase or decrease the specified index, in the right direction. */
2502 if (direction
== DIRECTION_DOWN
) {
2517 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2519 /* Consider it an error if any of the two offsets is uninitialized */
2520 if (old_offset
== 0 || new_offset
== 0)
2523 /* If we go down, the new offset must be larger than the old one. */
2524 return direction
== DIRECTION_DOWN
?
2525 new_offset
> old_offset
:
2526 new_offset
< old_offset
;
2529 int journal_file_next_entry(
2532 direction_t direction
,
2533 Object
**ret
, uint64_t *offset
) {
2541 n
= le64toh(f
->header
->n_entries
);
2546 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2548 r
= generic_array_bisect(f
,
2549 le64toh(f
->header
->entry_array_offset
),
2550 le64toh(f
->header
->n_entries
),
2559 r
= bump_array_index(&i
, direction
, n
);
2564 /* And jump to it */
2566 r
= generic_array_get(f
,
2567 le64toh(f
->header
->entry_array_offset
),
2575 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2576 * the next one might work for us instead. */
2577 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2579 r
= bump_array_index(&i
, direction
, n
);
2584 /* Ensure our array is properly ordered. */
2585 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2586 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2596 int journal_file_next_entry_for_data(
2598 Object
*o
, uint64_t p
,
2599 uint64_t data_offset
,
2600 direction_t direction
,
2601 Object
**ret
, uint64_t *offset
) {
2608 assert(p
> 0 || !o
);
2610 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2614 n
= le64toh(d
->data
.n_entries
);
2619 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2621 if (o
->object
.type
!= OBJECT_ENTRY
)
2624 r
= generic_array_bisect_plus_one(f
,
2625 le64toh(d
->data
.entry_offset
),
2626 le64toh(d
->data
.entry_array_offset
),
2627 le64toh(d
->data
.n_entries
),
2637 r
= bump_array_index(&i
, direction
, n
);
2643 r
= generic_array_get_plus_one(f
,
2644 le64toh(d
->data
.entry_offset
),
2645 le64toh(d
->data
.entry_array_offset
),
2653 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2655 r
= bump_array_index(&i
, direction
, n
);
2660 /* Ensure our array is properly ordered. */
2661 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2662 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2672 int journal_file_move_to_entry_by_offset_for_data(
2674 uint64_t data_offset
,
2676 direction_t direction
,
2677 Object
**ret
, uint64_t *offset
) {
2684 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2688 return generic_array_bisect_plus_one(f
,
2689 le64toh(d
->data
.entry_offset
),
2690 le64toh(d
->data
.entry_array_offset
),
2691 le64toh(d
->data
.n_entries
),
2698 int journal_file_move_to_entry_by_monotonic_for_data(
2700 uint64_t data_offset
,
2703 direction_t direction
,
2704 Object
**ret
, uint64_t *offset
) {
2712 /* First, seek by time */
2713 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2719 r
= generic_array_bisect_plus_one(f
,
2720 le64toh(o
->data
.entry_offset
),
2721 le64toh(o
->data
.entry_array_offset
),
2722 le64toh(o
->data
.n_entries
),
2724 test_object_monotonic
,
2730 /* And now, continue seeking until we find an entry that
2731 * exists in both bisection arrays */
2737 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2741 r
= generic_array_bisect_plus_one(f
,
2742 le64toh(d
->data
.entry_offset
),
2743 le64toh(d
->data
.entry_array_offset
),
2744 le64toh(d
->data
.n_entries
),
2752 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2756 r
= generic_array_bisect_plus_one(f
,
2757 le64toh(o
->data
.entry_offset
),
2758 le64toh(o
->data
.entry_array_offset
),
2759 le64toh(o
->data
.n_entries
),
2781 int journal_file_move_to_entry_by_seqnum_for_data(
2783 uint64_t data_offset
,
2785 direction_t direction
,
2786 Object
**ret
, uint64_t *offset
) {
2793 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2797 return generic_array_bisect_plus_one(f
,
2798 le64toh(d
->data
.entry_offset
),
2799 le64toh(d
->data
.entry_array_offset
),
2800 le64toh(d
->data
.n_entries
),
2807 int journal_file_move_to_entry_by_realtime_for_data(
2809 uint64_t data_offset
,
2811 direction_t direction
,
2812 Object
**ret
, uint64_t *offset
) {
2819 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2823 return generic_array_bisect_plus_one(f
,
2824 le64toh(d
->data
.entry_offset
),
2825 le64toh(d
->data
.entry_array_offset
),
2826 le64toh(d
->data
.n_entries
),
2828 test_object_realtime
,
2833 void journal_file_dump(JournalFile
*f
) {
2841 journal_file_print_header(f
);
2843 p
= le64toh(f
->header
->header_size
);
2845 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2849 switch (o
->object
.type
) {
2852 printf("Type: OBJECT_UNUSED\n");
2856 printf("Type: OBJECT_DATA\n");
2860 printf("Type: OBJECT_FIELD\n");
2864 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
2865 le64toh(o
->entry
.seqnum
),
2866 le64toh(o
->entry
.monotonic
),
2867 le64toh(o
->entry
.realtime
));
2870 case OBJECT_FIELD_HASH_TABLE
:
2871 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2874 case OBJECT_DATA_HASH_TABLE
:
2875 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2878 case OBJECT_ENTRY_ARRAY
:
2879 printf("Type: OBJECT_ENTRY_ARRAY\n");
2883 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
2884 le64toh(o
->tag
.seqnum
),
2885 le64toh(o
->tag
.epoch
));
2889 printf("Type: unknown (%i)\n", o
->object
.type
);
2893 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
2894 printf("Flags: %s\n",
2895 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
2897 if (p
== le64toh(f
->header
->tail_object_offset
))
2900 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
2905 log_error("File corrupt");
2908 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
2911 x
= format_timestamp(buf
, l
, t
);
2917 void journal_file_print_header(JournalFile
*f
) {
2918 char a
[33], b
[33], c
[33], d
[33];
2919 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
2921 char bytes
[FORMAT_BYTES_MAX
];
2926 printf("File Path: %s\n"
2930 "Sequential Number ID: %s\n"
2932 "Compatible Flags:%s%s\n"
2933 "Incompatible Flags:%s%s%s\n"
2934 "Header size: %"PRIu64
"\n"
2935 "Arena size: %"PRIu64
"\n"
2936 "Data Hash Table Size: %"PRIu64
"\n"
2937 "Field Hash Table Size: %"PRIu64
"\n"
2938 "Rotate Suggested: %s\n"
2939 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2940 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2941 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
2942 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
2943 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
2944 "Objects: %"PRIu64
"\n"
2945 "Entry Objects: %"PRIu64
"\n",
2947 sd_id128_to_string(f
->header
->file_id
, a
),
2948 sd_id128_to_string(f
->header
->machine_id
, b
),
2949 sd_id128_to_string(f
->header
->boot_id
, c
),
2950 sd_id128_to_string(f
->header
->seqnum_id
, d
),
2951 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
2952 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
2953 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
2954 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
2955 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
2956 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
2957 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
2958 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
2959 le64toh(f
->header
->header_size
),
2960 le64toh(f
->header
->arena_size
),
2961 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
2962 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
2963 yes_no(journal_file_rotate_suggested(f
, 0)),
2964 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
2965 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
2966 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
2967 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
2968 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
2969 le64toh(f
->header
->n_objects
),
2970 le64toh(f
->header
->n_entries
));
2972 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
2973 printf("Data Objects: %"PRIu64
"\n"
2974 "Data Hash Table Fill: %.1f%%\n",
2975 le64toh(f
->header
->n_data
),
2976 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
2978 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
2979 printf("Field Objects: %"PRIu64
"\n"
2980 "Field Hash Table Fill: %.1f%%\n",
2981 le64toh(f
->header
->n_fields
),
2982 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
2984 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
2985 printf("Tag Objects: %"PRIu64
"\n",
2986 le64toh(f
->header
->n_tags
));
2987 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2988 printf("Entry Array Objects: %"PRIu64
"\n",
2989 le64toh(f
->header
->n_entry_arrays
));
2991 if (fstat(f
->fd
, &st
) >= 0)
2992 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
2995 static int journal_file_warn_btrfs(JournalFile
*f
) {
3001 /* Before we write anything, check if the COW logic is turned
3002 * off on btrfs. Given our write pattern that is quite
3003 * unfriendly to COW file systems this should greatly improve
3004 * performance on COW file systems, such as btrfs, at the
3005 * expense of data integrity features (which shouldn't be too
3006 * bad, given that we do our own checksumming). */
3008 r
= btrfs_is_filesystem(f
->fd
);
3010 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3014 r
= read_attr_fd(f
->fd
, &attrs
);
3016 return log_warning_errno(r
, "Failed to read file attributes: %m");
3018 if (attrs
& FS_NOCOW_FL
) {
3019 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3023 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3024 "This is likely to slow down journal access substantially, please consider turning "
3025 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3030 int journal_file_open(
3037 JournalMetrics
*metrics
,
3038 MMapCache
*mmap_cache
,
3039 Set
*deferred_closes
,
3040 JournalFile
*template,
3041 JournalFile
**ret
) {
3043 bool newly_created
= false;
3049 assert(fd
>= 0 || fname
);
3051 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
3052 (flags
& O_ACCMODE
) != O_RDWR
)
3056 if (!endswith(fname
, ".journal") &&
3057 !endswith(fname
, ".journal~"))
3061 f
= new0(JournalFile
, 1);
3069 f
->prot
= prot_from_flags(flags
);
3070 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3071 #if defined(HAVE_LZ4)
3072 f
->compress_lz4
= compress
;
3073 #elif defined(HAVE_XZ)
3074 f
->compress_xz
= compress
;
3081 f
->mmap
= mmap_cache_ref(mmap_cache
);
3083 f
->mmap
= mmap_cache_new();
3091 f
->path
= strdup(fname
);
3092 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3093 asprintf(&f
->path
, "/proc/self/%i", fd
);
3099 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3100 if (!f
->chain_cache
) {
3106 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3112 /* fds we opened here by us should also be closed by us. */
3116 r
= journal_file_fstat(f
);
3120 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3122 (void) journal_file_warn_btrfs(f
);
3124 /* Let's attach the creation time to the journal file,
3125 * so that the vacuuming code knows the age of this
3126 * file even if the file might end up corrupted one
3127 * day... Ideally we'd just use the creation time many
3128 * file systems maintain for each file, but there is
3129 * currently no usable API to query this, hence let's
3130 * emulate this via extended attributes. If extended
3131 * attributes are not supported we'll just skip this,
3132 * and rely solely on mtime/atime/ctime of the file. */
3134 fd_setcrtime(f
->fd
, 0);
3137 /* Try to load the FSPRG state, and if we can't, then
3138 * just don't do sealing */
3140 r
= journal_file_fss_load(f
);
3146 r
= journal_file_init_header(f
, template);
3150 r
= journal_file_fstat(f
);
3154 newly_created
= true;
3157 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3162 r
= mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3168 if (!newly_created
) {
3169 if (deferred_closes
)
3170 journal_file_close_set(deferred_closes
);
3172 r
= journal_file_verify_header(f
);
3178 if (!newly_created
&& f
->writable
) {
3179 r
= journal_file_fss_load(f
);
3187 journal_default_metrics(metrics
, f
->fd
);
3188 f
->metrics
= *metrics
;
3189 } else if (template)
3190 f
->metrics
= template->metrics
;
3192 r
= journal_file_refresh_header(f
);
3198 r
= journal_file_hmac_setup(f
);
3203 if (newly_created
) {
3204 r
= journal_file_setup_field_hash_table(f
);
3208 r
= journal_file_setup_data_hash_table(f
);
3213 r
= journal_file_append_first_tag(f
);
3219 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
)) {
3224 if (template && template->post_change_timer
) {
3225 r
= journal_file_enable_post_change_timer(
3227 sd_event_source_get_event(template->post_change_timer
),
3228 template->post_change_timer_period
);
3234 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3241 if (f
->fd
>= 0 && mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
3244 (void) journal_file_close(f
);
3249 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3250 _cleanup_free_
char *p
= NULL
;
3252 JournalFile
*old_file
, *new_file
= NULL
;
3260 if (!old_file
->writable
)
3263 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3264 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3265 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3268 if (!endswith(old_file
->path
, ".journal"))
3271 l
= strlen(old_file
->path
);
3272 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3273 (int) l
- 8, old_file
->path
,
3274 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3275 le64toh((*f
)->header
->head_entry_seqnum
),
3276 le64toh((*f
)->header
->head_entry_realtime
));
3280 /* Try to rename the file to the archived version. If the file
3281 * already was deleted, we'll get ENOENT, let's ignore that
3283 r
= rename(old_file
->path
, p
);
3284 if (r
< 0 && errno
!= ENOENT
)
3287 /* Sync the rename to disk */
3288 (void) fsync_directory_of_file(old_file
->fd
);
3290 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3291 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3292 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3293 * would result in the rotated journal never getting fsync() called before closing.
3294 * Now we simply queue the archive state by setting an archive bit, leaving the state
3295 * as STATE_ONLINE so proper offlining occurs. */
3296 old_file
->archive
= true;
3298 /* Currently, btrfs is not very good with out write patterns
3299 * and fragments heavily. Let's defrag our journal files when
3300 * we archive them */
3301 old_file
->defrag_on_close
= true;
3303 r
= journal_file_open(-1, old_file
->path
, old_file
->flags
, old_file
->mode
, compress
, seal
, NULL
, old_file
->mmap
, deferred_closes
, old_file
, &new_file
);
3305 if (deferred_closes
&&
3306 set_put(deferred_closes
, old_file
) >= 0)
3307 (void) journal_file_set_offline(old_file
, false);
3309 (void) journal_file_close(old_file
);
3315 int journal_file_open_reliably(
3321 JournalMetrics
*metrics
,
3322 MMapCache
*mmap_cache
,
3323 Set
*deferred_closes
,
3324 JournalFile
*template,
3325 JournalFile
**ret
) {
3329 _cleanup_free_
char *p
= NULL
;
3331 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3333 -EBADMSG
, /* corrupted */
3334 -ENODATA
, /* truncated */
3335 -EHOSTDOWN
, /* other machine */
3336 -EPROTONOSUPPORT
, /* incompatible feature */
3337 -EBUSY
, /* unclean shutdown */
3338 -ESHUTDOWN
, /* already archived */
3339 -EIO
, /* IO error, including SIGBUS on mmap */
3340 -EIDRM
, /* File has been deleted */
3341 -ETXTBSY
)) /* File is from the future */
3344 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3347 if (!(flags
& O_CREAT
))
3350 if (!endswith(fname
, ".journal"))
3353 /* The file is corrupted. Rotate it away and try it again (but only once) */
3356 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3358 now(CLOCK_REALTIME
),
3362 if (rename(fname
, p
) < 0)
3365 /* btrfs doesn't cope well with our write pattern and
3366 * fragments heavily. Let's defrag all files we rotate */
3368 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3369 (void) btrfs_defrag(p
);
3371 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3373 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3376 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3378 uint64_t q
, xor_hash
= 0;
3391 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3392 ts
.realtime
= le64toh(o
->entry
.realtime
);
3394 n
= journal_file_entry_n_items(o
);
3395 /* alloca() can't take 0, hence let's allocate at least one */
3396 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3398 for (i
= 0; i
< n
; i
++) {
3405 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3406 le_hash
= o
->entry
.items
[i
].hash
;
3408 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3412 if (le_hash
!= o
->data
.hash
)
3415 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3418 /* We hit the limit on 32bit machines */
3419 if ((uint64_t) t
!= l
)
3422 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3423 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3426 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3427 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3431 data
= from
->compress_buffer
;
3434 return -EPROTONOSUPPORT
;
3437 data
= o
->data
.payload
;
3439 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3443 xor_hash
^= le64toh(u
->data
.hash
);
3444 items
[i
].object_offset
= htole64(h
);
3445 items
[i
].hash
= u
->data
.hash
;
3447 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3452 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3454 if (mmap_cache_got_sigbus(to
->mmap
, to
->fd
))
3460 void journal_reset_metrics(JournalMetrics
*m
) {
3463 /* Set everything to "pick automatic values". */
3465 *m
= (JournalMetrics
) {
3466 .min_use
= (uint64_t) -1,
3467 .max_use
= (uint64_t) -1,
3468 .min_size
= (uint64_t) -1,
3469 .max_size
= (uint64_t) -1,
3470 .keep_free
= (uint64_t) -1,
3471 .n_max_files
= (uint64_t) -1,
3475 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3476 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3483 if (fstatvfs(fd
, &ss
) >= 0)
3484 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3486 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3490 if (m
->max_use
== (uint64_t) -1) {
3493 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3495 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3496 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3498 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3499 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3501 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3503 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3505 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3506 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3509 if (m
->min_use
== (uint64_t) -1)
3510 m
->min_use
= DEFAULT_MIN_USE
;
3512 if (m
->min_use
> m
->max_use
)
3513 m
->min_use
= m
->max_use
;
3515 if (m
->max_size
== (uint64_t) -1) {
3516 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3518 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3519 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3521 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3523 if (m
->max_size
!= 0) {
3524 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3525 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3527 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3528 m
->max_use
= m
->max_size
*2;
3531 if (m
->min_size
== (uint64_t) -1)
3532 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3534 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3536 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3537 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3539 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3540 m
->max_size
= m
->min_size
;
3543 if (m
->keep_free
== (uint64_t) -1) {
3546 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3548 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3549 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3552 m
->keep_free
= DEFAULT_KEEP_FREE
;
3555 if (m
->n_max_files
== (uint64_t) -1)
3556 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3558 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3559 format_bytes(a
, sizeof(a
), m
->min_use
),
3560 format_bytes(b
, sizeof(b
), m
->max_use
),
3561 format_bytes(c
, sizeof(c
), m
->max_size
),
3562 format_bytes(d
, sizeof(d
), m
->min_size
),
3563 format_bytes(e
, sizeof(e
), m
->keep_free
),
3567 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3573 if (f
->header
->head_entry_realtime
== 0)
3576 *from
= le64toh(f
->header
->head_entry_realtime
);
3580 if (f
->header
->tail_entry_realtime
== 0)
3583 *to
= le64toh(f
->header
->tail_entry_realtime
);
3589 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3597 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3601 if (le64toh(o
->data
.n_entries
) <= 0)
3605 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3609 *from
= le64toh(o
->entry
.monotonic
);
3613 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3617 r
= generic_array_get_plus_one(f
,
3618 le64toh(o
->data
.entry_offset
),
3619 le64toh(o
->data
.entry_array_offset
),
3620 le64toh(o
->data
.n_entries
)-1,
3625 *to
= le64toh(o
->entry
.monotonic
);
3631 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3635 /* If we gained new header fields we gained new features,
3636 * hence suggest a rotation */
3637 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3638 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3642 /* Let's check if the hash tables grew over a certain fill
3643 * level (75%, borrowing this value from Java's hash table
3644 * implementation), and if so suggest a rotation. To calculate
3645 * the fill level we need the n_data field, which only exists
3646 * in newer versions. */
3648 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3649 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3650 log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items, %llu file size, %"PRIu64
" bytes per hash table item), suggesting rotation.",
3652 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3653 le64toh(f
->header
->n_data
),
3654 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3655 (unsigned long long) f
->last_stat
.st_size
,
3656 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3660 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3661 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3662 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3664 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3665 le64toh(f
->header
->n_fields
),
3666 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3670 /* Are the data objects properly indexed by field objects? */
3671 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3672 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3673 le64toh(f
->header
->n_data
) > 0 &&
3674 le64toh(f
->header
->n_fields
) == 0)
3677 if (max_file_usec
> 0) {
3680 h
= le64toh(f
->header
->head_entry_realtime
);
3681 t
= now(CLOCK_REALTIME
);
3683 if (h
> 0 && t
> h
+ max_file_usec
)